This is an adaptation of @fwinkl's !55759 (merged) to generic
AthenaMP. In a nutshell, this gives the possibility of executing certain algorithms/sequences during
PreFork incident in
AthenaMP. One can use this to execute the conditions algorithms before the first event, hence sharing the memory across workers even without FAFE.
Based on an 8 process, 10k event
AODtoDAOD test w/
SharedWriter, vanilla MP uses about 2 GB/worker, executing conditions algorithms before the event loop this way reduces this to 1.5 GB/core, while FAFE still performs the best at 0.9 GB/worker:
We can take a look to see if there is any milage in adopting such an approach in other jobs such as derivation production.